Skills Management in Unique AI Conduct

Overview

Skills are reusable capability packages that extend the Conduct agent's behavior for specific workflows. Each skill is a folder in the Knowledge Base containing a SKILL.md manifest (with YAML front matter defining name and description) and optional supporting files — scripts, templates, reference documents, or configuration.

When a Conduct space is configured with a skills scope, the agent downloads all skills in that scope at the start of every session. The agent then uses matching skills automatically based on the user's task.

This article covers how to organize, manage, and create skills for your organization.

Who is it for?

Administrators and team leads responsible for configuring Conduct spaces and managing the skills library for their organization.

If you're unable to access certain features or sections of this article, it's possible that your firm doesn't have access or hasn't upgraded to the latest version. Please reach out to your internal support team for further assistance.

Who should have access

• Skills Admins — manage the company-wide skills library

• Team leads — manage team-specific skills

• Use-case owners — manage skills for specific workflows

• Knowledge Base write access is required to create or edit skills

Skill Anatomy

Every skill is a folder containing at minimum a SKILL.md file:

my-skill/
├── SKILL.md           ← Required — manifest with instructions
├── scripts/           ← Optional — executable code for deterministic tasks
├── references/        ← Optional — docs loaded into context as needed
├── assets/            ← Optional — templates, icons, fonts
└── examples/          ← Optional — input/output examples

SKILL.md Format

---
name: "financial-document-parser"
description: "Extracts key data points, tables, and metadata from financial documents (10-K, DDQ, side letters). Use when the user uploads a financial PDF or asks to parse/extract from a regulatory document."
---

[Detailed instructions for the agent — what to extract, how to format output,
edge cases to handle, quality checks to run]

The name and description in the front matter are critical — the description determines when the agent selects this skill. Make it specific about trigger conditions.

Organizing Skills

We recommend organizing skills in the Knowledge Base by scope, mirroring how permissions and ownership typically flow in regulated environments:

Skills/
├── out-of-the-box/           [read: all]   [write: Unique]
│   ├── doc-coauthoring
│   ├── pdf
│   ├── docx
│   ├── pptx
│   ├── xlsx
│   ├── screenshot
│   ├── data-analysis
│   └── chart-generator
│
├── company/                  [read: all]   [write: Skills Admins]
│   ├── house-style-memo
│   ├── bloomberg-ticker-resolver
│   └── compliance-disclaimer-injector
│
├── team/                     [read: team]  [write: team leads]
│   ├── long-short-equity/
│   ├── credit/
│   ├── macro/
│   └── risk-and-operations/
│
├── use-case/                 [read: scoped] [write: use-case owner]
│   ├── ddq-rfp/
│   ├── 10-K-analysis/
│   ├── earnings-call/
│   ├── side-letter-extraction/
│   └── portfolio-reconciliation/
│
└── personal/                 [read+write: owner only]
    ├── alex.albracht/
    ├── pm.smith/
    └── analyst.chen/

Skill Graduation

Skills naturally graduate upward as they prove out:

1. Personal — Drafted by an individual user for their own workflow

2. Team — Adopted by a desk or team after proving useful

3. Company — Standardized into the company-wide library once trusted and requested for broad adoption

4. Out-of-the-box — Maintained by Unique for all customers

Baseline Skills

Unique provides a set of baseline skills covering the highest-frequency workflows across our financial services client base. These are split between general-purpose and financial services-specific:

General Skills (provided by Unique)

Financial Services Skills (provided by Unique)

Company-Specific Skills

We recommend developing company-specific skills where:

• Intellectual property is involved (proprietary models, analysis frameworks)

• Secure system access is required (internal APIs, databases)

• House style needs enforcement (memo formats, branding, disclaimers)

• Regulatory compliance demands specific output structures

Configuring Skills for a Space

Step 1: Create a Skills Folder in the Knowledge Base

1. Navigate to the Knowledge Base in the admin interface

2. Create a folder structure following the organization pattern above

3. Upload skill folders (each containing a SKILL.md and supporting files)

Step 2: Set the Skills Scope on the Space

1. Open the Conduct space in the admin interface

2. The skills scope is configured via the skills_scope_id parameter, which points to a Knowledge Base folder

3. All skills within that folder (and subfolders) are downloaded to the agent at session start

Step 3: Verify Skills Are Loading

1. Open the space as a user

2. Ask the agent: "What skills do you have available?"

3. The agent should list the skills loaded from the configured scope

Writing Effective Skills

The Description Matters Most

The description field in SKILL.md front matter is the primary trigger for skill selection. The agent reads it to decide whether to use the skill. Make it specific:

Good: "Extracts key data points, tables, and metadata from financial documents (10-K, DDQ, side letters). Use when the user uploads a financial PDF or asks to parse/extract from a regulatory document."

Bad: "Processes documents"

Instruction Patterns

• Explain the why — Tell the agent why each instruction matters, not just what to do. This helps it make judgment calls on edge cases.

• Include examples — Show input/output pairs so the agent understands the expected format.

• Define output structure — Specify the exact template or format the output should follow.

• Handle errors — Describe what to do when data is missing, malformed, or ambiguous.

• Keep it focused — One skill per workflow. Don't bundle unrelated capabilities.

Bundled Scripts

For deterministic or repetitive tasks, include executable scripts in scripts/. The agent can run these directly instead of writing code from scratch — faster and more reliable.

Testing Skills

We recommend an iterative testing loop:

1. Configure the full set of tools, MCP servers, sub-agents, and skills the agent will need in production

2. Run representative tasks end-to-end in the UI — observe tool selection, skill triggering, and output quality

3. Refine skill descriptions and instructions based on where the agent hesitated, misrouted, or produced poor output

4. Repeat via the API for automated regression once behavior stabilizes

Skill descriptions and instructions are the highest-leverage thing to tune — they determine whether the right skill fires and how well it executes.

FAQs

Q: How many skills can a space have?
A: There's no hard limit on the number of skills, but keep in mind that all skills are downloaded at session start. A very large number of skills increases setup time and can dilute the agent's ability to select the right one. Focus on quality over quantity.

Q: Can different spaces use different skills?
A: Yes — each space has its own skills_scope_id pointing to a Knowledge Base folder. You can configure different skill sets for different teams, use cases, or client deployments.

Q: How do I update a skill?
A: Edit the skill files in the Knowledge Base. The next time a user starts a session in the space, they'll get the updated version automatically. No space reconfiguration needed.

Q: Can users override or disable a skill?
A: Not currently. Skills loaded for a space apply to all users of that space. If a user doesn't want a skill's behavior, they can instruct the agent explicitly (e.g., "don't use a template — give me raw output").

Q: What happens if two skills overlap?
A: The agent uses the skill whose description best matches the user's task. If descriptions overlap, the agent may choose unpredictably. Keep descriptions distinct and specific to avoid conflicts.

Q: Can skills access the Knowledge Base or external APIs?
A: Skills themselves are static packages (instructions + files). However, the agent can use skills alongside MCP tools — so a skill's instructions can tell the agent to search the Knowledge Base or call an MCP server as part of its workflow.

Calling the Harness via API

The Conduct harness is fully compatible with the Unique API — it behaves like any other space. You authenticate the same way, send messages to the same completion endpoint, and stream responses identically. Tool calls, skill invocations, and sub-agent handoffs are visible in the response stream.

Spaces are configured (connected tools, MCP servers, sub-agents, skills) in the admin interface, and all API calls against that space inherit the space configuration.

Security

• Skills are stored in the Knowledge Base and subject to the same access controls as any KB content

• Skills are downloaded over the platform's internal APIs — no external network access required

• Skill content is read-only inside the agent's environment

• Skills cannot escalate the agent's permissions — they can only guide how the agent uses its existing tools

Author

Alex Albracht